Kling AI image generation

Turn the creative request into one precise, user-confirmed remote image generation. Follow the Claude Code OAuth, billing confirmation, single-submission, status, and result rules in the core Kling Skill.

Read prompt construction for reference images, exact copy, branded assets, or controlled variants. Read scene patterns for product, advertising, portrait, poster, or cover work. For Element reuse or local media, also read the core asset workflows.

Workflow

1. Use text_to_image for a new image. Use image_to_image when source media controls identity, product structure, composition, editable content, or style. An attachment alone does not determine the mode.
2. Ask only for missing information that materially changes the result: subject, destination and aspect ratio, required copy, reference roles, protected facts, or output count.
3. Create or reuse the UUIDv7 taskTraceId for this objective, call who_am_i, and select an explicitly compatible model and live arguments.
4. Upload local images first and use only the live input names and returned URLs. Never silently degrade to text-to-image after an upload failure.
5. Build one prompt covering subject, action, environment, composition, lighting, color, materials, camera language, protected facts, and necessary exclusions.
6. Show the final mode, model, prompt summary, resolution, aspect ratio, output count, and reference roles. Obtain explicit approval under the core Skill's billing rule; honor an existing explicit authorization for the final settings.
7. Call the image generation tool exactly once and preserve the task number. If its MCP App mounts, let that one App self-refresh and do not call query_tasks from the model. If no App mounts, follow the core Skill's headless polling fallback; never resubmit.
8. When Claude Code renders the MCP App, do not add Markdown media or duplicate links. When no App is rendered, provide one primary image link.

Mode rules